HAL Id: hal-00478453 https://hal.archives-ouvertes.fr/hal-00478453 Submitted on 30 Apr 2010 HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. Mood and gradability: an investigation of the subjunctive mood in Spanish Elisabeth Villalta To cite this version: Elisabeth Villalta. Mood and gradability: an investigation of the subjunctive mood in Spanish. Linguistics and Philosophy, Springer Verlag, 2009, 31 (4), pp.467-522. 10.1007/s10988-008-9046-x. hal-00478453
82
Embed
Mood and gradability: an investigation of the subjunctive ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
HAL Id: hal-00478453https://hal.archives-ouvertes.fr/hal-00478453
Submitted on 30 Apr 2010
HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.
Mood and gradability: an investigation of thesubjunctive mood in Spanish
Elisabeth Villalta
To cite this version:Elisabeth Villalta. Mood and gradability: an investigation of the subjunctive mood in Spanish.Linguistics and Philosophy, Springer Verlag, 2009, 31 (4), pp.467-522. �10.1007/s10988-008-9046-x�.�hal-00478453�
and many more. Most of these proposals have concentrated on characterizing the common properties
of the contexts that trigger the subjunctive mood rather than spelling out an explicit semantics for the
predicates that select the subjunctive mood. For reasons of space, I will not go into the details of these
approaches here.1 I will limit myself to comparing my proposal to those approaches that present an
explicit proposal for the semantics of these predicates, such as Heim (1992), von Fintel (1999) and
Giorgi and Pianesi (1997). In particular, Giorgi and Piannesi (1997) propose that the licensing
contexts for subjunctive mood can be characterized with Kratzer=s semantics for modality (cf.
Kratzer 1977, 1981, 1999). I will argue that an approach which uses such a semantics for these
predicates cannot account for a number of important properties (entailment relations, practical
inferences, and contexts with more than two alternatives). To the contrary, the proposal developed
here accommodates these properties in a straightforward way.
Finally, in this paper, new empirical evidence for two crucial properties of the predicates that
select the subjunctive mood will be discussed: these predicates are focus sensitive (cf. Dretske 1972)
and they are gradable. These two properties follow directly from the proposal developed here. In the
vast literature on mood, the link between the appearance of the subjunctive mood and these important
properties has never been made before.
The structure of this paper is as follows. In section 2, the relevant data are presented. In
section 3, Heim=s(1992) semantics for desire predicates and emotive factive predicates is presented.
In section 4, a new semantics for these predicates is developed based on the comparison of contextual
alternatives. In section 5, the proposal is extended to all predicate classes that select the subjunctive
mood in Spanish. In section 6, this proposal is compared to previous proposals for the semantics of
these predicates (Giorgi and Pianesi1997, von Fintel1999). In section 7, I investigate the focus
sensitivity of these predicates (cf. Dretske 1972,1975) and make a proposal for the semantic
contribution of the subjunctive mood morpheme. Finally, in section 8, I present empirical evidence for
the gradable nature of these predicates and refine my analysis: predicates that select the subjunctive
mood are analyzed as having an extra degree argument.
Error! Main Document Only. See Portner (1999, 2003), as well as Villalta (2006) for an extensive discussion of traditional and recent approaches to the semantics of mood.
4
2. The distribution of subjunctive mood in Spanish complement clauses
In Spanish, one important factor that determines whether the verb of an embedded clause is in
the indicative or the subjunctive mood is the matrix predicate. The indicative mood is selected by
aconsejar >advise=, suggerir >suggest=) and causatives (e.g., hacer >make=, conseguir >achieve=).
A few examples are presented below.
(5) Victoria quiere que Marcela venga al picnic. TO WANT
Victoria wants that Marcela come:PRES.SUBJ.3SG to-the picnic.
>Victoria wants Marcela to come to the picnic.=
(6) Sofía duda que Rafael pueda venir. TO DOUBT
Sofia doubts that Rafael can:PRES.SUBJ.3SG come.
>Sofía doubts that Rafael can come.=
(7) Marcela se alegra de que la hayan invitado. TO BE GLAD
Marcela SE glad of that PRO her have:PAST.SUBJ.3PL invited.
>Marcela is glad that they have invited her.=
(8) Victoria sugiere que salgan temprano. TO SUGGEST
Victoria suggests that PRO leave:PRES.SUBJ.3PL early.
>Victoria suggests that they leave early.=
It can be noted that, in both categories, we find factive as well as non-factive predicates.
Sentences that contain factive predicates have the presupposition that the proposition expressed by the
complement clause is true, contrary to sentences with non-factive predicates. Among the predicates
that select the indicative mood, we find factive predicates such as saber (>know=), acordarse
(>remember=) and olvidarse (>forget=), as well as non-factives such as creer (>believe=), prometer
(>promise=), decir (>say=) and soñar (>dream=). Among the predicates that select the subjunctive
mood, all emotive factive predicates and causative predicates are factive.
A number of predicates allow both the indicative and the subjunctive mood in their
complement clause. Crucially, however, mood alternation then correlates with a meaning change in
the predicate. Depending on the mood of the complement clause, the predicates fall under the
corresponding predicate classes as described above. A few examples are presented below.
6
The predicate sentir can either be interpreted as an emotive factive predicate (>be sorry=) and
then selects the subjunctive mood, as in (9), or it can be interpreted as a predicate of perception
(>sense=/=have the impression=) and then selects the indicative mood, as in (10).
(9) Siento que te hayan hecho daño.
PRO sorry that PRO you have:PAST.SUBJ.3PL done pain.
>I am sorry that they have hurt you.=
(10) Siento que va a haber un problema.
PRO sense that PRO go:FUT.IND.3SG to there-be a problem.
>I have the impression that there is going to be a problem.=
The predicate decir can either be interpreted as a predicate of communication (>tell=/=say=)
and then selects the indicative mood, as in (11), or it can be interpreted as a directive predicate
(>order=) and then selects the subjunctive mood, as in (12). Most of the predicates of communication
are ambiguous in this sense.
(11) Te digo que acabaré a tiempo.
PRO you tell that PRO finish:FUT.IND.1SG on time.
>I tell you that I will finish on time.=
(12) Te dije que acabaras a tiempo.
PRO you told that PRO finish:PAST.SUBJ.2SG on time.
>I told you to finish on time.=
Thus, when a predicate can select both the indicative and the subjunctive mood, the alternation
correlates with a meaning change of the predicate. Notice, crucially, that the meaning change does not
always correlate with a factive/non-factive distinction.
In the following, I turn to a first attempt at characterizing the semantics of the predicates
that select the subjunctive mood in Spanish.
7
3. A semantics for the predicates that select the subjunctive mood in Spanish
In this section, the goal is to develop a first proposal for a common semantics of the predicates
that select the subjunctive mood in Spanish. I first present Heim=s (1992) proposal for desire
predicates. Heim=s (1992) analysis of desire predicates builds on Stalnaker=s (1984) insight that
every desire report contains a hidden conditional. AA little more explicitly, the leading intuition is that
John wants you to leave means that John thinks that if you leave he will be in a more desirable world
than if you don=t.@ (Heim 1992, p.193). I then argue that such a conditional semantics should be
extended to all predicates that select the subjunctive mood in Spanish.2
3.1 Heim=s (1992) conditional semantics for desire predicates
We begin with the semantics for propositional attitude predicates first introduced by Hintikka
(1969). The truth conditions for the verb believe are given in (13).
(13) >α believes φ= is true in w iff: � w=0 Doxα(w): φ is true in w=.
(Doxα(w) contains all the worlds that are compatible with what α believes in the world w to be
true, also called doxastic alternatives of α in w.)
The truth conditions in (13) say that >α believes φ= is true in a world w if and only if φ is true
in all worlds w= that are compatible with what α believes in w. Thus, to take an example, John
believes that it is raining is true in w iff it is raining in all the worlds w= that are compatible with
what John believes in w.
Adopting the Hintikka-style analysis for a verb such as want results in the truth conditions
given in (14). Here, the worlds compatible with α=s beliefs in w, the doxastic alternatives, have
simply been replaced with the worlds compatible with α=s desires in w, the so-called >buletic=
alternatives.
Error! Main Document Only. Giannakidou (1998,1999), Portner (1999, 2003) and Quer (1998, 2001) also point out that the conditional semantics proposed by Heim (1992) is relevant for mood selection. But they do not discuss the consequences of adopting such a semantics.
8
(14) >α wants φ= is true in w iff: � w=0 Bulα(w): φ is true in w=.
(Bulα(w) contains all the worlds that are compatible with what α desires in w, also called
buletic alternatives of α in w.)
These truth conditions, however, cannot appropriately capture all the characteristics of the
verb want. This becomes clear as soon as we look at an example more closely. The example in (15),
from Heim (1992), illustrates that these truth conditions are problematic.
(15) I want to teach Tuesdays and Thursdays next semester.
Heim notes that the sentence in (15) can be true in a situation in which in fact I would rather
prefer not to teach at all. In a situation like this the following holds: in all the worlds that are
compatible with what I desire I do not teach. If we adopt the truth conditions in (14), however, the
sentence comes out as false in this situation: the proposition is only evaluated in worlds that are
compatible with my desires, which do not include worlds in which I teach Tuesdays and Thursdays
next semester.
To remedy this problem, Heim (1992) develops a new proposal for want that captures
Stalnaker=s insight that every desire report employs a hidden conditional: >to want something, is to
prefer something to certain relevant alternatives, the relevant alternatives being those possibilities that
the agent believes will be realizable if he doesn=t get what he wants.= (Stalnaker 1984, p.89). The
sentence (15) can thus be paraphrased as in (16).
(16) If I teach Tuesdays and Thursdays next semester, I will be in a more desirable world than if I
teach on other days next semester.
Heim=s new proposal for the truth conditions of want is based on the semantics adopted for
conditionals. Following Lewis (1973) and Stalnaker (1968), a conditional if φ, ψ is true in a world w
iff ψ is true in all φ-worlds maximally similar to w (where a φ-world is a world in which φ is true).
Thus, in (16), the sentence is true if and only if the consequent is true in all those worlds in which I
teach Tuesday and Thursdays next semester and which are otherwise just like the actual world. By
9
adopting such a conditional semantics, Heim proposes that the verb want has the truth-conditions in
(17).
(17) >α wants φ= is true in w iff:
For every w=0 Doxα(w): Every φ-world maximally similar to w= is more desirable to α in w
than any non-φ-world maximally similar to w=.
(Heim 1992, p.193)
We now apply this proposal to our previous example repeated here in (18).
(18) I want to teach Tuesdays and Thursdays next semester.
Under the truth conditions presented above, for this sentence to be true, the following has to
hold: for each doxastic alternative w=, if I teach Tuesdays and Thursdays next semester in w=, it is a
more desirable world than maximally similar worlds in which I teach on other days. If I teach on other
days in w=, it is a less desirable world than maximally similar worlds in which I do teach on Tuesdays
and Thursdays.
In Heim (1992), the truth conditions of this verb are stated in a context change semantics.
Since I do not use this framework here, I present the equivalent truth conditions in a non-dynamic
semantics3. Heim encodes the relation of comparative similarity among worlds with a family of
selection functions as in (19): for each world w, there is a selection function Simw from propositions
to propositions which maps each proposition p to the set of worlds maximally similar to w in which p
is true (cf. Heim 1992, p.195).
(19) Simw(p)={w=0 W: w=0 p and w= resembles w no less than any other world in p}
3 By using a dynamic framework, Heim solves a number of presupposition projection facts that will remain unaddressed here.
10
Additionally, Heim uses an abbreviation for the ranking of possible worlds in terms of
desirability. She introduces >α,w4, a relation between worlds, that can also be employed in an extended
sense as a relation between sets of worlds, as defined in (20) (see Heim 1992, p.197).
(20) (a) For any w, w=, w== 0 W, w= >α,w w== iff w= is more desirable to α in w than w==
(b) For any w 0 W, XφW, YφW, X >α,wY iff w= >α,w w== for all w=0 X, w==0 Y.
The truth conditions of the verb want can now be stated as in (21).
(21) � want� (p)(a)(w) = 1 iff
� w= 0 Doxa(w): Simw=(p) >a,w Simw=(5p)
Heim then suggests that this conditional semantics can be extended to other predicates such as
wish and be glad. She argues, however, that there is an important difference between these predicates:
in the case of the predicate want all the desirability comparisons should be entirely among worlds that
the subject believes possible, contrary to predicates such as wish and be glad.
Heim thus proposes to further restrict the truth conditions of want. The truth conditions in (21)
only require that all p-worlds that are maximally similar to w= be more desirable than all 5 p-worlds
maximally similar to w=. Nothing is said about whether these maximally similar worlds have to be
doxastic alternatives or not. In the new proposal, in (22), all desirability comparisons are entirely
among the subject=s belief worlds, because the argument of the similarity function Sim applies to a
subset of the doxastic alternatives. The function Sim now returns the set of worlds that are an element
of Doxa(w) 1 p and Doxa(w) 1 5p respectively and which resemble w no less than any other world.
Error! Main Document Only. To express that X is more desirable than Y, I will use X >α,w Y rather than X <α,w Y as is originally done in Heim (1992). The definition is otherwise identical to the one in Heim(1992).
11
It is exactly in this aspect that the predicate want differs from a predicate such as wish. This is best
illustrated with an example.
(23) John wishes he taught on Tuesdays and Thursdays.
This sentence can only be judged to be true in a situation in which John believes that he
doesn=t teach on Tuesdays and Thursdays. Hence, the worlds compatible with his beliefs only include
worlds in which he doesn=t teach on Tuesdays and Thursdays. Thus, we cannot adopt the same truth
conditions as for want. If all desirability comparisons are entirely among the subject=s belief worlds,
the set on the left hand side of the comparison relation will always be empty. Heim proposes that, for
the verb wish, the proposition p is evaluated with respect to a revision of the worlds compatible with
his beliefs, a set that will also include some p-worlds. Heim proposes that this revised set of worlds
results from Doxa(w) by suspending some of the assumptions in Doxa(w), as in the revision of a
context necessary for the interpretation of counterfactuals. The definition proposed in Heim (1992) is
stated in a context change semantics and is not suitable for the semantics adopted here. I will assume
here that the revision of the doxastic alternatives with respect to a proposition p, which I call
revp(Doxa(w)), contains all the worlds in Doxa(w) as well as all the p-worlds most similar to w.
We can now adopt the truth conditions of the predicate wish given in (24). Here, in the first
argument of the ordering relation >a,w, the function Sim applies to worlds that are included in the
To summarize, in Heim (1992), predicates such as wish, want and be glad share their core
semantics, a conditional semantics. They have the common feature that for each doxastic alternative,
two sets of worlds are compared. The only difference between these predicates lies in whether the
relevant sets of worlds are included in the doxastic alternatives of the subject or rather in a revision of
the doxastic alternatives of the subject.
3.2 A conditional semantics for predicates that select the subjunctive mood
After this brief presentation of Heim=s semantics for desire predicates, we now return to the
class of predicates that select the subjunctive mood in Spanish. The predicates discussed by Heim
(1992) all select the subjunctive mood in Spanish. My hypothesis will be that the conditional
semantics adopted so far can also be extended to all other predicate classes that select the subjunctive
mood in Spanish, such as modals, predicates of doubt, directives and causatives. This hypothesis
naturally accommodates the fact that among the predicates that select the subjunctive mood we find
factive as well as non-factive predicates. The semantic property shared by these predicates is that they
establish a comparison. Whether this comparison is between worlds that are among the doxastic
alternatives of the subject or not is irrelevant for mood selection.
Under this hypothesis, the predicates that select the subjunctive mood in Spanish share the
following common property: they introduce an ordering relation or scale. It is important to observe
that the ordering relation expressed by the predicate is not a relation of desirability for all predicates.
For instance, predicates such as sorprenderse >be surprised= and dudar >doubt= rather require an
ordering relation of likelihood. Also, a number of emotive factive predicates cannot be based on the
13
notion of desirability, such as es interesante (>it is interesting=), es divertido (>it is amusing=), etc. I
will assume that the ordering relation is contributed by the lexical meaning of each predicate. This is
reminiscent of what has been said about the meaning of gradable adjectives in the literature. The
scalar approaches to the semantics of gradable adjectives argue that the meaning of a gradable
adjective such as tall directly contributes the dimension of the scale involved (in the case of tall the
dimension is height).5
Below, I present a first attempt at a characterization of the predicates that select the
subjunctive mood in Spanish, as stated in (27).
(27) Preliminary hypothesis for the subjunctive mood in Spanish
A proposition p that is the complement of the matrix predicate requires the subjunctive mood
iff the matrix predicate introduces an ordering relation and compares p to non-p.
To conclude, I have proposed to adopt Heim=s (1992) conditional semantics to characterize
the predicates that select the subjunctive mood in Spanish. In the following, I present two arguments
that call for a revision of this semantics.
4. A new proposal: comparison of contextual alternatives
In this section, I propose to revise the semantics for the predicates that select the subjunctive
mood in Spanish. First, I argue that their semantics involves comparison of the embedded proposition
p with contextually available alternatives, rather than simply with 5p. Second, I argue that, in the truth
conditions, reference to the subject=s beliefs can be replaced with the set of contextually available
alternatives. I then develop a new analysis for these predicates.
Error! Main Document Only. For approaches that adopt a scalar semantics for gradable adjectives see Bartsch and Venneman 1973, Bierwisch 1989, Cresswell 1976, Kamp 1975, Kennedy 1999, Klein E. 1980, 1991, Rullmann 1995, Schwarzschild & Wilkinson 1999, Seuren 1973, von Stechow 1984a,1984b, and many others.
14
4.1 When more than two contextual alternatives are available
The central point to be made in this section is that the semantics of the predicates selecting the
subjunctive mood requires comparison of contextual alternative propositions. In fact, the semantics
adopted so far is based on a special case where the context only provides the alternatives p and 5p.
Contexts providing more than two alternatives illustrate this issue quite clearly. Consider the scenario
in (28).
(28) Sofía has promised to bring a dessert to the picnic. Victoria believes that there are three
possibilities for what she may actually do. She could prepare a chocolate cake, even though
Victoria considers that extremely unlikely because it represents far too much work. She might
bring an apple pie, which Victoria considers very likely since she can just buy it at the bakery
nearby. Or Sofía might bring ice-cream, which seems most likely to Victoria, since she
usually has some in her freezer. Victoria prefers the chocolate cake over the apple pie and the
apple pie over the ice-cream.
The schematic figure in (29) represents the two scales involved in this scenario, desirability
according to Victoria and likelihood according to Victoria. That Sofía will bring the chocolate cake is
the most desirable and most unlikely alternative; that she will bring the ice cream is the least desirable
and most likely alternative.
(29) + desirable ┬ chocolate cake ┬ - likely (according to Victoria) │ │ │ │ │ │ │ apple pie │ │ │
! desirable ┴ ice cream ┴ + likely
In this situation, the sentence in (30) is intuitively judged to be false.
(30) Victoria wishes Sofía would bring an apple pie.
15
Contrary to intuitions, Heim=s conditional semantics for the verb wish predicts that this
sentence should come out as true in this scenario. Under Heim=s proposal, the sentence (30) is true iff
for all worlds w= that are compatible with Victoria=s beliefs, the worlds in which Sofía brings the
apple pie are more desirable to Victoria than all the minimally different worlds in which she doesn=t.
This is true, since, in the given scenario, the closest worlds in which Sofía doesn=t bring an apple pie
are worlds in which she brings ice-cream. This set does not contain worlds in which she brings
chocolate cake (these are not minimally different for Victoria since she considers them as extremely
unlikely). A conditional semantics thus does not make the correct predictions for this scenario.
Other predicates give rise to similar effects in contexts that make more than two alternatives
available. Imagine that, in the continuation of the previous scenario, Sofía does in fact bring the apple
pie to the picnic. The sentence in (31) can then be true in that situation.
(31) Victoria is disappointed that Sofía brought an apple pie.
A conditional semantics for the predicate be disappointed, however, predicts (31) to be false
in this scenario. Simplifying somewhat, I will assume that the only difference between the semantics
of be disappointed and be glad is that the ordering relation of desirability is reversed. The truth
conditions of be disappointed then predict that the worlds in which Sofía brought the apple pie are
less desirable to Victoria than all the minimally different worlds in which she doesn=t. This is false,
since the closest worlds are all worlds in which she brings ice-cream. She considers these worlds as
less desirable. Again, a conditional semantics does not make the correct predictions for a scenario as
described above.
I conclude that the semantics of the predicates under discussion involves comparison of p with
the set of its contextual alternatives rather than with just 5p. I propose to adopt a new lexical entry for
want as in (32). The difference with the previous lexical entry for want is that, here 5p is replaced by
the contextually given alternatives q.
(32) Semantics of want based on comparison of alternatives
According to (34), from the truth of (33a) it follows that in DoxI(w) all the worlds in which I
teach Tuesdays and Thursdays next semester are more desirable than maximally similar worlds in
which I don=t teach on Tuesdays and Thursdays next semester. From the truth of (33b) it follows that
in DoxI(w) the worlds in which I teach Tuesdays and Thursdays next semester correspond exactly to
those worlds in which I work hard this semester. Hence, from (33a) and (33b) it follows that in
DoxI(w) all worlds in which I work hard this semester are more desirable than maximally similar
worlds in which I don=t work hard this semester. As a result, the inference is expected to be valid,
and (33c) is expected to be true, contrary to what we observe.
The invalid inference illustrates that someone may believe that two propositions p and q are
true in the exact same set of worlds, and at the same time want p without wanting q. Heim=s
semantics cannot capture this fact. As a consequence, I propose to eliminate the reference to the
doxastic alternatives and simply replace it with the set of contextually relevant alternatives.
The proposal made in (32) can then be simplified accordingly, as in (35) along with a slight
modification of the definition of >α,w6:
(35) Semantics of want based on comparison of alternatives
� wantC� g(p)(a)(w) = 1 iff
� q: q � p & q 0 g(C): p >DESα,w q
where >DESα,w is defined as follows
a) For any w, w=, w== 0 W, w= > α,w w== iff w= is more desirable to α in w than w==.
b) For any pφW, qφW, p >DES α,w q iff � w== 0 q � w=0 p such that w=>α,w w==, and it is
not the case that � w= 0 p � w==0 q such that w==>α,w w=.
Error! Main Document Only. In (35), I propose to use a new definition of >DESα,w, adapted from the definition of >better possibility= in Kratzer(1991), since the ordering relation defined in Heim (1992) is not suitable for comparative desirability among propositions. This is so, since Heim=s definition makes a requirement that is too strong. If we were to apply it to propositions we would run into the following problem: if p is more desirable than q, the definition requires that all worlds in p be more desirable than all worlds in q. However, some worlds in p may be really bad worlds for other reasons, and may still not necessarily affect the desirability relation between p and q. Kratzer=s definition of >better possibility= solves this problem.
18
With the semantics proposed in (35), no inference is expected. The new semantics predicts
that >I teach Tuesdays and Thursdays= is more desirable than its contextually relevant alternatives
(for example >I teach Mondays, Wednesdays and Fridays=). However, from the fact that >I teach
Tuesdays and Thursdays next semester= and >I work hard this semester= are true in the exact same
set of worlds it does not follow that >I want to work hard this semester=. This is so, since in both
cases we are comparing a different set of alternatives. For example, the set of alternatives for >I work
had this semester= may be {>I don=t work this semester=, >I work a little this semester=}. As a
consequence, no inference is expected.
With the semantics proposed here, however, two new problems arise. First, we cannot account
anymore for the fact that the subject=s beliefs do sometimes play a role in determining the truth
conditions of a want-sentence. Take the example in (36).
(36) John doesn=t have to teach at all next semester but he thinks he does and he wants to teach
Tuesdays and Thursdays.
In this sentence, the alternative AJohn doesn=t teach at all next semester@ is provided in the context,
however, it cannot be part of the set of alternatives that is relevant for the comparison. If this were so,
we would not predict the correct truth conditions. Hence, the subject=s beliefs do play a role in
determining which alternatives enter the comparison.
A second problem is that we cannot express the difference between want, wish and be glad
anymore. As we saw before, these predicates precisely differ in terms of whether the subject believes
that the proposition expressed by the complement clause is true or not. As a consequence, I propose to
distinguish these predicates in terms of their definedness conditions.
Below, I adopt a semantics in which the predicate want has a definedness condition which
requires that all contextual alternatives be included in the doxastic alternatives7.
(37) Semantics of want based on comparison of alternatives
� wantC� g(p)(a)(w) is defined iff � q 0 g(C): Doxa(w)1q � ι
7 Further refinements may be necessary. As an anonymous reviewer points out, the predicate want requires that p be a desirable alternative, unlike the predicate prefer. That is, in a context in which Victoria considers all alternatives as highly undesirable it is still possible to say that Victoria prefers p, but not that Victoria wants p.
19
if defined � wantC� g(p)(a)(w) = 1 iff
� q: q � p & q 0 g(C): p >DESα,w q
I will show that this new proposal still does not predict the practical inference discussed earlier
to be valid. The practical inference is repeated here.
(38) a) I want to teach Tuesdays and Thursdays next semester.
b) I believe that I will teach Tuesdays and Thursdays next semester if and only if I work hard
now.
____________________________________
c) Invalid inference:
� I want to work hard now.
According to (37), from the truth of (38a) it follows that >I teach Tuesdays and Thursdays
next semester= is more desirable to me than its contextual alternatives, for example >I teach
Mondays, Wednesdays and Fridays next semester=. From the truth of (38b) it follows that in DoxI(w)
the worlds in which I teach Tuesdays and Thursdays next semester correspond exactly to those worlds
in which I work hard this semester. From this, however, it does not follow that (38c) is true. (38c)
says that >I work hard now= is more desirable than its contextual alternatives, for instance {>I don=t
work at all=, >I work a little=}. Even if the propositions >I teach Tuesdays and Thursdays next
semester= and >I work hard now= are true in the same set of worlds, their contextual alternatives are
not necessarily the same.
I propose that the predicates wish and be glad will then have slightly different definedness
conditions. The fact that the predicate wish requires that the proposition p should not be included in
the subject=s doxastic alternatives, whereas the predicate be glad requires that the proposition p
necessarily be true in the subject=s doxastic alternatives can now be expressed in their definedness
conditions.
(39) Semantics of wish based on comparison of alternatives
� wishC� g(p)(a)(w) is defined iff p 1 Doxa(w) = ι
If defined � wishC� g(p)(a)(w) = 1 iff
� q: q� p & q 0 g(C): p >DESα,w q
20
(40) Semantics of be glad based on comparison of alternatives
� be gladC� g(p)(a)(w) is defined iff Doxa(w) φ p
If defined � be gladC� g(p)(a)(w) = 1 iff
� q: q� p & q 0 g(C): p >DESα,w q
The lexical entries provided here are certainly a simplification, given that the precise
definedness conditions of these predicates are a rather complex matter (cf. von Fintel 1992). Since my
main aim is to show that these different predicates share a core semantics, I will not go into further
discussion of the details of these definedness conditions here, but adopt a simplified semantics that is
sufficient for the argumentation developed here.
I will now propose that the semantics based on comparison of contextual alternatives can be
extended to all predicates that select the subjunctive mood in Spanish:
(41) New hypothesis for the subjunctive mood in Spanish
A proposition p that is the complement of the matrix predicate requires the subjunctive mood
iff the matrix predicate introduces an ordering relation between propositions and compares p
to its contextually available alternatives.
To conclude, in this section, I have developed a new hypothesis for the semantics of the
predicates that select the subjunctive mood in Spanish. By revising Heim=s (1992) proposal, I have
developed a semantics of comparison for predicates such as want, be glad and wish. The new
semantics involves comparison of contextual alternatives to the proposition expressed by the
complement clause. In the next section, I discuss how the presented analysis can be extended to all
predicate classes that select the subjunctive mood.
5. The predicate classes that select the subjunctive mood in Spanish
In this section, I will go over all predicate classes that select the subjunctive mood in Spanish
to show how their semantics can be captured by the generalization. While it is impossible to present a
21
full fledged analysis for every single predicate class here, I will briefly sketch possible ways of
analyzing the different classes under the presented proposal. The exact semantics of causative
predicates will however be left for future research.
5.1. Desire predicates and Emotive Factive Predicates
In the previous section, I have discussed the desire predicates want and wish and emotive
factive predicates such as be glad and be disappointed. The analysis can in principle be extended to
all other predicates in these two classes, but there are two important factors that may vary.
The first factor of variation is the dimension of the scale which may differ from predicate to
predicate. This scale should be contributed directly by the lexical meaning of the predicate. Desire
predicates such as querer (>want=) and esperar (>hope=) contribute an ordering relation of
desirability. For the class of emotive factive predicates there is more variation in terms of what
dimension is contributed. Many emotive factive predicates contribute a relation of desirability, such
as alegrarse (>be glad=) and estar contento (>be happy=), or a reverse relation of desirability, such
as lamentarse (>regret=), or enfadarse (>be upset=). However, a number of other emotive factive
predicates cannot be based on the notion of desirability such as es interesante (>it is interesting=), es
divertido (>it is amusing=), es útil (>it is useful=), es fascinante (>it is fascinating=), etc. I will
assume that the ordering relation is contributed by the lexical meaning of each predicate. Parallel to
the scalar approaches to the semantics of gradable adjectives, I will propose that gradable predicates
directly contribute the dimension of the scale involved.
The second factor of variation is the ranking of p among the alternatives. The proposal
developed here leaves room for expressing that p may be the best, the worst or just a good alternative.
While I have argued here that these predicates require p to be the best (or highest ranked) among the
contextual alternatives, we will see in section 6.2.3 that not all predicates require that p be the best
alternative, but rather that p be a good alternative (such as, for example, alegrarse >be glad= or
lamentarse >be disappointed=).
5.2 Modals and predicates expressing likelihood and doubt
22
Modal predicates, which select the subjunctive mood in Spanish, can be captured by the
generalization quite straightforwardly. I will illustrate this here with the predicate es necesario >be
necessary=. Krasikova (2007) argues that have to-like modals such as have to, need, be required, be
necessary etc. are associated with the likelihood scale/scale of comparative possibility when they
express teleological modality.8 Parallel to Krasikova=s (2007) analysis, I will define the lexical entry
for the predicate be necessary as follows:
(42) Semantics of be necessary based on comparison of alternatives
� be necessaryC� g(p)(w) = 1 iff � q: q� p & q 0 g(C): p >LIKELYw q
where >LIKELYw is defined as follows:
a) For any w, w=, w== 0 W, w= > w w== iff w= is closer to w than w==.
b) For any pφW, qφW, p >LIKELYw q iff � w== 0 q � w=0 p such that w=>αw w==, and it
is not the case that � w= 0 p � w==0 q such that w==>,w w=.
The likelihood scale as defined here is based on comparative similarity (closeness to the actual world,
cf. Lewis1986). Other predicates from this class such as es probable (>it is likely=), es posible (>it is
possible=) and dudar (>doubt=) can also be analyzed as contributing the scale of likelihood. There
may again be variation concerning the ranking of p among the contextual alternatives. For example, in
the case of es posible (>it is possible=), the requirement will be that p is not the best alternative.
Here as well, there will be variation in the kind of scale contributed by the predicate: when
modals express other kinds of modalities (such as deontic or bouletic modalities), the scale
contributed is a different one (such as ranking based on the closeness to the ideal or law, cf. Lewis
1986).
8 She argues this on the basis of the fact that their ability to induce the more than minimum reading in comparatives correlates to their ability to appear in sufficiency modal constructions (cf. Von Fintel & Iatridou 2007).
23
Sloman (1970) in fact already proposes an analysis equivalent to the one presented here for
sentences containing the predicate ought.9 In order to capture their various possible interpretations
(modal can, must and directive interpretation), he develops an analysis in which alternatives are
compared with respect to a contextually determined ordering relation. He proposes that the
interpretation of It ought to be the case that p should be >p is, or is a necessary condition for the best,
relative to the basis B, of the possibilities in the class Z.= (Sloman 1970, p. 389). B here stands for
basis of comparison (dimension of the ordering relation), which is contextually given when not
specified explicitly. Z stands for the comparison class. Sloman claims that possible different meanings
of ought fall out of differences in basis of comparison as well as differences in comparison class.
5.3. Directive predicates
Portner(2004,2007) proposes that imperative sentences are used to contribute to the discourse
component that he labels the To-do List, more precisely the addressee=s To-do List. A To-do List is a
set of propositions that the participant intends to make true. In short, imperative sentences are used to
make requirements, where the essential function of a requirement is to add a proposition to
someone=s To-Do List. Just as declarative sentences are used to make assertions and contribute to the
common ground, in Portner (2004,2007), imperatives are used to make requirements and contribute to
the addressee=s To-Do List. Another important feature of his analysis is that the To-Do List is
considered to be a subset of the ordering source (cf. Kratzer 1981) and thus contributes to imposing an
ordering on the common ground.
Turning now to directive predicates such as mandar (>order=), ordenar (>order=), pedir
(>ask=), we can show that, by adopting Portner=s main ideas, these can be accommodated in the
proposal developed here. I propose that they make reference to the To-Do List and that the relevant
set of propositions are those propositions that are part of (or a subset of) the set of contextual
alternatives. The ordering relation is introduced by the predicate. For deontic predicates, such as to
order, I propose the following semantic definition:
(43) Semantics of to order based on comparison of alternatives
9 I thank Kai von Fintel for pointing this out to me.
24
� orderC� g(p)(x)(w) = 1 iff x requires that
� q: q� p & q 0 g(C) & q 0 T(α) : p >DEONTICw q
where T(α) is the set of propositions (>To-Do List=) assigned to a
participant α in the conversation
The ordering relation contributed by the predicate may vary. In Portner(2007), the ordering imposed
may vary depending on what is expressed by the imperative: when the imperative expresses an order,
it will be a deontic ordering source, when it expresses an invitation, it will express a bouletic ordering
source, when it expresses a suggestion, it will express a teleological ordering source.
Notice that under this proposal the semantics of directives is slightly different from the other
predicate classes that select the subjunctive mood: the comparative meaning component is embedded
more deeply within the meaning of the predicate.
5.4 Causative predicates
In the case of causatives such as hacer (>make=) and lograr (>achieve=), I propose that the
relevant comparison relation is one of comparative similarity. Causatives are usually analyzed with
the semantics of counterfactual conditionals. In Lewis (1973), it is proposed that, in a counterfactual,
all worlds in which the antecedent is true are ordered with respect to their similarity to the actual
world. For a causative predicate this means that, all worlds in which the embedded proposition is true
will be compared to the actual world with respect to their similarity. This ordering relation may also
be defined for propositions (cf Lewis 1986) as was done for modal predicates in section 5.2. Hence,
the predicate cause can be defined as expressing that p is the closest to the actual world among the set
of contextual alternatives.
I suggest that predicates such as hacer (>make=) and lograr (>achieve=) contain the
predicate cause as part of their meaning. Hence, similarly to directive predicates, the relevant
comparison will be embedded more deeply within the meaning of the predicate. Even though I will
not go into the details of a semantics of causatives here, we can assume that they share with the other
predicates classes the fact that they contribute an ordering relation and compare p its contextual
alternatives.
25
6. Comparing the proposal to other approaches
In this section, I compare my proposal to other existing proposals on the semantics of
propositional attitude predicates.
Propositional attitude predicates have received a lot of interest in the semantic and
philosophical literature independently of the issues of mood selection. Since Hintikka (1962, 1969),
predicates such as believe, know and want have commonly been analyzed with a semantics of possible
worlds, more specifically with a semantics of modal necessity: α believes/knows/wants φ is true in w
iff φ is true in all the worlds that are compatible with α=s beliefs/knowledge/desires in w. The
question then is to what extent a Hintikka-style semantics could be extended to all kinds of
propositional attitudes. In a similar vein, recent approaches have proposed that propositional attitudes
should be modeled after the semantics of modal expressions involving a semantics of necessity and
possibility (cf., for example Kiefer 1987, von Fintel 1999, Giorgi and Piannesi 1997).
In this section, I explicitly compare my proposal to these approaches. As we have already seen
before, a Hintikka-style semantics has its problems when extended to desire predicates. I will show
here that there are a number of characteristics of the predicates that select the subjunctive mood that
are difficult to account for in this type of approach. We will see that, on the contrary, these
characteristics follow straightforwardly from the proposal that I have adopted here.
I now turn to making explicit how the meaning common to the predicates that select the
subjunctive mood can be expressed with a semantics of modal necessity or modal possibility,
following von Fintel(1999). Since he makes use of the semantics proposed in Kratzer (1997, 1981,
1999) to capture the meaning properties of predicates such as want, wish, be glad and be sorry, I first
present a brief overview of Kratzer=s theory of modality.
6.1 A semantics of modal necessity/possibility
Kratzer (1977, 1981, 1991) develops a semantics in which modal expressions receive
meanings that are dependent on two contextual parameters, a modal base and an ordering source.
These two parameters are determined by two conversational backgrounds.
26
Kratzer points out that there are many different kinds of conversational backgrounds. A
conversational background is the kind of entity denoted by expressions such as what the law provides,
what we know, what is normal, what is rational, what is desirable, etc. What the law provides in a
world is a set of propositions p such that the law provides that p in that world. The denotation of what
the law provides is then the function that assigns to every possible world this set of propositions p.
More generally, a conversational background is a function f which assigns sets of propositions to
possible worlds.
The modal base is defined as a set of worlds determined by a conversational background f.
The set of worlds in which all propositions of f(w) are true constitutes the modal base in w. The
modal base thus determines the set of worlds accessible from each world.
The ordering source is a second conversational background g which assigns to every possible
world a set of propositions. Ordering sources represent ideals, given that they can induce an ordering
on the modal base (the worlds in the modal base may be closer or further away from what the law
provides, what is desirable, what is normal, etc.).
More generally, a set of propositions A can induce an ordering #A on a set of worlds W in the
following way (following Lewis 1981).
(44) The ordering #A:
� w , w= 0 W: w #A w= iff {p: p0 A and w=0 p} φ {p: p0 A and w0 p}
>A world w is at least as close to the ideal represented by A as a world w= iff all propositions
of A which are true in w= are true in w as well.=
Thus, in (44), worlds are ordered with the help of an unordered set of propositions. A world w is at
least as close to the ideal represented by A as a world w= iff it makes at least as many propositions
true as w= does.
Sentences containing modals are then evaluated with respect to an ordered set of worlds. The
ordering source is usually not expressed explicitly in the sentence but has to be recovered from the
context. Consider the following example in (45).
(45) Sofía should bring a chocolate cake.
27
In the context of the scenario discussed previously, in section 4.1, the modal base of this sentence is
formed by the set of worlds that constitute Victoria=s beliefs. The ordering source is bouletic and
corresponds to the set of propositions determined by Victoria=s desires. This set of propositions
induces an ordering on the set of the modal base. The modal force of the modal should is necessity.
The meaning of (45) can then be paraphrased as in (46).
(46) In view of what Victoria desires, it is necessary for Sofía to bring a chocolate cake.
We now turn to the implementation of a semantics based on modal necessity for predicates
that select the subjunctive mood.
Giorgi and Pianesi (1997) propose that the contexts in which the subjunctive mood appear can
be characterized with the semantics that Kratzer proposes for modal expressions. The generalization
they propose is that in certain Romance languages such as Spanish and French, the subjunctive mood
is selected when the context requires the interaction between a modal base and a non-null ordering
source. This generalization still does not tell us what exactly the semantics should be for the
predicates that select the subjunctive mood. Giorgi and Pianesi do not explicitly formalize a proposal
for the semantics of these predicates, but only illustrate the kind of semantics they want to use with
the modal must. I will thus follow here instead von Fintel (1999), who proposes to adopt a semantics
of modal necessity for propositional attitude predicates such as want, wish, be glad and be sorry. He
defines the following concepts.
(47) (i) The >modal base function= f is a function from pairs of an individual and a world to a set
of worlds.
(ii) The >ordering source function= is a function from pairs of an individual and a world to a
set of propositions (in the case of want to a set of propositions forming the subject=s
preferences).
(von Fintel 1999, p.115)
Here, the subject=s preferences form an unordered set of propositions that induce an ordering on the
modal base. The best worlds are those that make most propositions true. This is stated in the
following definition.
28
(48) For any set of propositions P, we define a strict partial order <P:
� w=,w==: (w= <P w== iff � p0 P (w== 0 p 6 w=0p) & � p 0 P (w=0 p & w==⌠p))
(Von Fintel 1999, p. 115)
Von Fintel proposes to use the selection function maxP that selects the best worlds in any set X with
respect to a partial order <P, as defined in (49).
(49) For a given strict partial order <P on worlds, define the selection function maxP that selects the
set of <P-best worlds from any set X:
� x φ W: maxP(X) = {w 0 X: 5� w=0 X : w= <P w}
(von Fintel 1999, p.116)
The semantics of the verb want can then be stated as in (50), as proposed in von Fintel (1999).
The proposition is here evaluated in those worlds in the modal base f(α,w) that maximally satisfy the
preferences given by the ordering source g(α,w). Thus, the ordering source is used to induce an
This semantics of the predicate be glad does not make the correct predictions for the above
scenario either. The meaning of be glad expressed in (64) predicts that the example (62) should be
false in the given scenario. The lexical entry requires that there is a world among the worlds that best
match Victoria=s desires such that Sofía brings an apple pie in that world. Nevertheless, we already
saw before that this is false: since Victoria prefers the chocolate cake, all worlds that maximally
correspond to her preferences are worlds in which Sofía doesn=t bring an apple pie, but rather a
chocolate cake.
Hence, we have seen that neither modal necessity nor modal possibility can capture the fact
that, for certain predicates, p may not necessarily be the best alternative. The problem with this kind
37
of semantics is that we only consider worlds that best match the subject=s desires. But these may
sometimes in fact be irrelevant.
To the contrary, the proposal that I have developed so far can capture the truth conditions of a
predicate such as be glad, because it allows to express that p is not necessarily the best alternative.
The predicates under discussion seem to vary with respect to this issue. Some predicates, such as
prefer, for example, require p to be ranked highest among the alternatives, whereas other predicates
such as be glad and regret don=t. This information should then be encoded in the lexical entry of
these predicates. The lexical entry for be glad could be formalized as in (65).
(65) Semantics based on comparison of alternatives for be glad
� be gladC� g(p)(a)(w) = 1 iff � q: q � p & q 0 g(C): p > Desa,w q
This lexical entry for the predicate be glad does not require that p be the best alternative, but rather
that there be an alternative q such that p is more desirable than q. In this scenario, we predict the
sentence Victoria is glad that Sofía brought the apple pie to come out as true: there is an alternative
(>Sofía brings ice cream=) such that the alternative >Sofía brings an apple pie= is more desirable.
To conclude, we have seen that a semantics based on necessity or possibility cannot capture
the fact that certain predicates such as be glad do not require the proposition expressed by the
complement clause to be ranked highest. A semantics based on comparison of contextual alternatives
leaves room for variation in this domain. Different kinds of constraints on how the proposition p
stands with respect to the other alternatives can be implemented.
In Kratzer (1991), next to modal possibility and necessity, other modal strengths are also
defined, among which the modal strength of at least as good a possibility and better possibility. Let
me now show in what respect my proposal differs from Kratzer=s proposal on comparative
possibility, which is based on the interaction of a modal base with an ordering source:
(66) Definition 9 (Kratzer 1991:p644)
A proposition p is at least as good a possibility as a proposition q in a world w with respect
to a modal base f and an ordering source g iff for all u such that u 0 1 f(w) and u 0 q there is a
v 0 1 f(w) such that v#g(w) u and v 0 p.
38
(67) Definition 10 (Kratzer 1991: p.644)
A proposition p is a better possibility than a proposition q in a world w with respect to a
modal base f and an ordering source g iff p is at least as good a possibility as q and q is not at
least as good a possibility as p in w with respect to f and g.
The notion of better possibility is defined here for propositions p and q with respect to a modal base f
and an ordering source g. The ordering source corresponds to an unordered set of propositions that
induces an ordering on the relevant set of worlds in the modal base. Crucially, in this definition, the
propositions p and q to be compared are not part of the set of propositions that forms the ordering
source. Hence, we could not assume, as was previously done in von Fintel=s proposal for the
predicate want, that the ordering source contains those propositions that correspond to the subject=s
preferences (since, at least in the case of the predicate want, the proposition p and q to be compared
are presumably propositions that express the subject=s desires).
In my proposal, the set of contextual alternative propositions is not used to induce the
ordering. Rather, the different propositions expressing the desires of Victoria are ranked with respect
to an ordering relation or scale. Thus, the crucial difference between the two approaches is that in one
case the propositions that express the desires are unordered while they are ranked with respect to a
scale in the other.
As pointed out to me by Barbara Partee, in the approach that uses ordering sources, one may
express ranked desires by using a set of disjunctions as the relevant ordering source. For our scenario,
we would need the set {>Sofía brings a chocolate cake=, >Sofía brings a chocolate cake or apple
pie=, >Sofía brings a chocolate cake or apple pie or ice cream=}. Given such an ordering source, the
best worlds are then the worlds in which Sofía brings the chocolate cake: these are the only worlds
that make all three propositions true. If it seems plausible to use ordering sources of this kind, these
two approaches may turn out to be equivalent.
In the following section, however, I will show that the proposal developed here has two
advantages. Given that predicates are analyzed parallel to focus sensitive operators such as only, we
predict these predicates to be focus sensitive. In section 7.1, I show that this prediction is born out.
The Kratzer-style approach, which is based on the interaction of a modal base with an ordering
source, however, does not make this prediction. In section 7.2.1, I argue that the subjunctive mood
morpheme makes a particular semantic contribution, namely that it evaluates the contextual
39
alternatives for the matrix predicate. In the Kratzer-style semantics, there does not seem to be room
for such a semantic contribution: the subjunctive mood morpheme would simply have to make a
vacuous semantic contribution. Hence, the proposal developed here has the advantage of providing a
first step towards a better understanding of the question why subjunctive verb forms are obligatory
under the predicate classes discussed here.
7. Mood and Focus
In this section, I turn to investigating an important consequence of my proposal. Given the
analysis presented here, we expect the predicates that select the subjunctive mood to be focus
sensitive. The focus sensitivity of these predicates follows directly from the fact that they are analyzed
analogous to other focus sensitive operators such as only. We thus expect the meaning of these
predicates to be affected by a focused constituent in the embedded clause.
7.1. Focus sensitive predicates
As already pointed out in Dretske (1972,1975), focused phrases embedded under certain
propositional attitude predicates give rise to meaning differences. In the context of the previous picnic
scenario, the sentence in (68) is felicitous, while the sentence in (69) is not (words in capital are
focused).
(68) Victoria wants Sofia to bring A CHOCOLATE CAKE.
(69) Victoria wants SOFIA to bring a chocolate cake.
The proposal developed here can provide a rather straightforward analysis of the meaning
difference between examples (68) and (69). This theory makes explicit that for predicates such as
want an analogy to focus sensitive operators such as only is expected. Both want and only make
reference to a domain of quantification C of contextually determined alternatives, which following.
Rooth (1992) is constrained by focus.
In fact, Rooth (1985,1992) shows that focus has a truth conditional effect in the context of the
adverb only. He presents the following example: in a context in which Mary introduced Bill and Tom
40
to Sue and there were no other introductions, (70) is true, but (71) is false. These two sentences only
differ in terms of what constituent is focused (marked in capital letters).
(70) Mary only introduced Bill to SUE.
(71) Mary only introduced BILL to Sue.
For the predicates discussed here, we thus also predict differences in focus structure in the
embedded clause to have a truth conditional effect. In this section, my goal is to show that, while
predicates that select the subjunctive mood do show this effect, predicates that select the indicative
mood do not.
We begin with the class of desire predicates. I illustrate their behavior with the predicate want
and use some examples similar to the one discussed in Heim(1992). Consider the following context:
(72) In the linguistics department, at the faculty meeting, the teaching schedules of the different
faculty members for the upcoming semester are discussed. There is only one syntactician in
the department (John), one phonologist (Lisa), and two semanticists (Lara and Frank). John
can only teach syntax. Lara can teach syntax and semantics. There is some controversy on
which days John should teach his syntax classes. There are two options: he may teach syntax
on Tuesdays and Thursdays, or he may teach syntax on Mo, We & Fri. Lisa=s preferences are
the following: she would prefer it if Lara would teach syntax rather than John. But given that
John has to teach syntax, she prefers it if he teaches on Tuesdays and Thursdays rather than on
Mo, We & Fri (because she wants the teaching slot on Mo, We & Fri for her own phonology
class, which cannot conflict with the syntax class).
In this scenario, consider the following utterances:
(73) Lisa wants John to teach syntax ON TUESDAYS AND THURSDAYS.
(74) Lisa wants JOHN to teach syntax on Tuesdays and Thursdays.
41
In the described scenario, the utterance in (73) is true. However, the utterance in (74) cannot
be true (since Lisa would in fact prefer it if Lara would teach syntax, not John). The only difference
between these two utterances is that a different constituent is focused in the embedded clause.
We can thus conclude that, in the presence of the predicate want, the focus structure of the
embedded clause has a truth conditional effect. The same can be shown to hold for other predicates
that select the subjunctive mood in Spanish. For the class of emotive factive predicates, I illustrate
this with the predicate be glad.
In the context described above, assume that, at the end of the faculty meeting, it is decided that
John is indeed going to teach syntax on Tuesdays and Thursdays. Consider now the following
utterances.
(75) Lisa is glad that John teaches syntax ON TUESDAYS AND THURSDAYS.
(76) Lisa is glad that JOHN teaches syntax on Tuesdays and Thursdays.
Here again, (75) can be true in the described scenario, but not (76) (for the same reasons as before).
The predicate be glad can thus be considered to be focus sensitive.
Boer(1978) illustrates that this same kind of focus sensitivity also appears in conditional
sentences. He presents an example introduced by Dretske(1972) with the following scenario: Ted=s
father left a clause in his will stipulating that Ted can only receive his inheritance if he is married by a
certain date. In this context (77) is true, but (78) is false:
(77) If Ted hadn=t MARRIED Alice, he would have lost his inheritance.
(78) If Ted hadn=t married ALICE, he would have lost his inheritance.
Similarly, the causative predicate cause and the directive predicate demand10, which in Spanish select
the subjunctive mood, show focus sensitivity. Modeling our examples after the conditional sentences,
in the same context, the examples (79) and (81) are true, while (80) and (82) are false:
(79) His father caused Ted to MARRY Alice.
(80) His father caused Ted to marry ALICE.
10 See also Dretske(1975), p.415, for an example which illustrates that directive predicates such as >advise= are focus sensitive.
42
(81) His father demanded that Ted MARRY Alice.
(82) His father demanded that Ted marry ALICE.
Notice that a modal predicate such as be necessary also displays this pattern: in the same context, (83)
is true while (84) is false :
(83) In order to receive his inheritance, it was necessary that Ted MARRY Alice.
(84) In order to receive his inheritance, it was necessary that Ted marry ALICE.
We have thus seen examples for each predicate class that selects the subjunctive mood (desire
predicates, emotive factives, modals, directives and causatives) that show that the matrix predicate is
sensitive to the focus-structure of the embedded clause.
In contrast, consider now some examples containing predicates that select the indicative mood
in Spanish. The following examples contain the epistemic predicate know.
(85) Lisa knows that John teaches syntax ON TUESDAYS AND THURSDAYS.
(86) Lisa knows that JOHN teaches syntax on Tuesdays and Thursdays.
Contrary to the previous examples, here the matrix predicate does not seem to be sensitive to the
focus structure of the embedded clause. These examples are both true under the same circumstances:
all contexts that make one of them true make the other one true as well. The fact that the predicate
know is not focus sensitive can best be illustrated with a dialogue. If speaker A says: ALisa knows that
John teaches syntax ON TUESDAYS AND THURSDAYS.@, speaker B cannot reply ANo, that=s not
true. Lisa knows that JOHN teaches syntax on Tuesdays and Thursdays.@ without contradicting
himself. Notice that this is different for the predicate want. If speaker A says: ALisa wants John to
teach syntax ON TUESDAYS AND THURSDAYS.@, speaker B can than naturally reply ANo ,
that=s not true. Lisa wants JOHN to teach syntax on Tuesdays and Thursdays.@ without uttering a
contradiction11.
11 For some speakers, a more explicit context is necessary to make this a natural dialogue. For example speaker B could naturally reply: "well, that's not really true, as she doesn't mind which day these classes take place, as long as
43
Other predicates that select the indicative mood, such as predicates of communication are not
focus sensitive either, as can be illustrated with the predicate say:
(87) Lisa said that John teaches syntax ON TUESDAYS AND THURSDAYS.
(88) Lisa said that JOHN teaches syntax on Tuesdays and Thursdays.
In these examples, focus on a constituent in the embedded sentence creates alternatives that are not
evaluated by the predicate of the matrix clause: the alternatives that are available for an example such
as (87) are the alternatives to the whole clause, not the embedded clause: >Lisa said that John teaches
syntax on Tuesdays and Thursdays= and >Lisa said that John teaches syntax on Mondays,
Wednesdays and Fridays=. Similarly, in (88), the available alternatives are >Lisa said that John
teaches syntax on Tuesdays & Thursdays= and >Lisa said that Lara teaches syntax on Tuesdays and
Thursdays=.
It can easily be shown that the same holds for other predicate classes that select the indicative
mood in Spanish: predicates of certainty (e.g. estar convencido >be convinced=, estar seguro >be
certain=), commissives (prometer >promise=), and fiction predicates (e.g. soñar >dream=) have this
same property.
John is the one who does the teaching, so one should rather say that Lisa wants JOHN to teach syntax on tuesdays and thursdays." (personal communication, Jenny Doetjes).
In fact, in the literature, some examples have already been provided to show that focus does
not always have the same effect in the context of all propositional attitude predicates. The following
examples are from Boër (1979). He suggests that with predicates such as know and believe the
meaning differences induced by focused constituents in the complement clause are much less
perceptible.
(89) Tom knows/believes that Bob HIT Alice.
(90) Tom knows/believes that Bob hit ALICE.
Similarly, Dretske (1972) points out that predicates such as believe and say are not sensitive to
the contrastive differences of the embedded clause (see footnote 16, p.435).
44
Finally, let me illustrate with the example of the predicate sentir which can select both the
subjunctive and indicative mood, that such predicates are only focus sensitive when they select the
subjunctive mood. The predicate sentir, when it selects the subjunctive mood, has the meaning of the
emotive factive predicate >be sorry=. In the scenario above, (91) turns out to be false, while (92) is
true:
(91) Lisa siente que Juan de clases de sintaxis los MARTES Y JUEVES
>Lisa is sorry that Juan teaches syntax on TUE&THUR=
(92) Lisa siente que JUAN de clases de sintaxis los martes y jueves
>Lisa is sorry that JUAN teaches syntax on Tue&Thur=
To the contrary, when the predicate has the meaning of >sense=/=have the impression=, and selects
the indicative mood, the two examples (93) and (94) are truthconditionally equivalent:
(93) Lisa siente que Juan dará clases de sintaxis los MARTES Y JUEVES
>Lisa has the impression that Juan will teach syntax on TUE&THUR=
(94) Lisa siente que JUAN dará clases de sintaxis los martes y jueves
>Lisa has the impression that JUAN will teach syntax on Tue&Thur=
To conclude, the purpose of this section has been to argue that predicates that select the
subjunctive mood are sensitive to the presence of a focused constituent in the embedded clause: when
different constituents are focused, there is a truth conditional meaning difference. We thus have
evidence for the claim that the predicates that select the subjunctive mood are focus sensitive
operators.
In the following section, I turn to making precise what the semantic contribution of the
subjunctive mood morpheme is.
7.2 The semantic contribution of the subjunctive mood
45
In this section, I address the following question: what role does the subjunctive mood
morpheme play in the semantic composition of the sentence? The literature on the semantics of mood
rarely addresses this question. The main focus of the majority of previous studies has rather been to
provide a characterization of the contexts in which the subjunctive mood appears. The investigation of
what the subjunctive mood contributes to the meaning composition provides a first step towards a
better understanding of why subjunctive verb forms require special semantic licensing conditions and
why they are obligatory in certain contexts.
To begin, I introduce Rooth=s (1985,1992) terminology for the semantics of sentences
containing focused constituents and present his semantics for focus sensitive operators such as only.
7.2.1 Focus according to Rooth (1985,1992)
As mentioned before, Rooth (1985,1992) shows that focus has a truth conditional effect in the
context of the adverb only. I now present Rooth=s proposal for the meaning of sentences of the
following kind:
(95) Mary only introduced Bill to SUE.
(96) Mary only introduced BILL to Sue.
In Rooth (1985,1992), next to the ordinary semantic value, an additional semantic value is
used to express the contribution that focus makes to the meaning of a sentence. The focus semantic
value of a constituent α is represented as � α� ALT , its ordinary semantic value as � α� o. The focus
semantic value of a sentence is the set of propositions obtainable from the ordinary semantic value by
making a substitution in the position corresponding to the focused phrase. I present here the ordinary
and alternative semantic values for the sentences above without only.
(97) � Mary introduced Bill to [Sue]F� o = λw.Mary introduced Bill to Sue in w
(98) � Mary introduced Bill to [Sue]F� ALT ={λw.Mary introduced Bill to x in w/x 0 D}
(99) � Mary introduced [Bill]F to Sue� o = λw.Mary introduced Bill to Sue in w
(100) � Mary introduced [Bill]F to Sue� ALT ={λw.Mary introduced x to Sue/x 0 D}
46
The denotation of only carries an index C which (through the variable assignment g) refers to a
contextually determined set of propositions:
(101) � onlyC� go = λp.λw.� q: q 0 g(C) & q(w) 6 p = q
For a focus sensitive operator such as only, Rooth proposes that its domain of quantification
C is constrained by the focus semantic value. Focus introduces a variable C which anaphorically
constrains the domain of quantification of the focus sensitive operator. Rooth defines the ~-operator
which is adjoined to the variable C and evaluates the alternatives introduced by the focused
constituent. The sentence in (102) can be assumed to have the underlying representation in (103):
(102) Mary only introduced Bill to SUE.
(103) IP1 ri
onlyC IP2 ri
IP3 ~ C 6
Mary introduced Bill to [Sue]F
There are two constraints on the variable C introduced by focus interpretation. The first is that φ ~ C
presupposes that C is a subset of the focus semantic value for φ (and contains both the ordinary
semantic value of φ and an element distinct from the ordinary semantic value of φ). The choice of
antecedent for the variable C is free, but it is guided by the presuppositional constraint introduced by
the ~-operator:
ty (104) 4 φ ~ C 4o is defined only if � C� o φ � φ� ALT
ty If defined, 4 φ ~ C 4o = � φ� o
47
The second constraint is that, in the expression [φ ~ C], focus has been interpreted, so the semantic
effect of the foci in φ has to be neutralized:
ty
(105) 4 φ ~ C 4ALT = {� φ� o}
Rooth=s semantics can also be adopted for predicates such as want. Parallel to only, the
predicate want has been defined here as carrying an index C that refers to a contextually available set
of propositions. For convenience, I repeat my proposal for predicates such as want here:
(106) � wantC� g = λp.λx.λw.� q: q � p & q 0 g(C): p >DES x,w q
Just like in the case of only, the index C of the predicate want should be constrained by the
focus semantic value of the embedded clause. The ~-operator, if placed in the appropriate position in
the tree, evaluates the alternatives for the predicate. I illustrate this here with an example from the
previous section, with a focused constituent in the embedded clause:
(107) Lisa wants John to teach syntax ON TUESDAYS AND THURSDAYS.
The appropriate tree structure for the sentence in (107) is the one below, where ~C is attached at the
level of the CP of the embedded clause.
48
(108) IP ep
NP VP g eo
Lisa V CP1 g ri wantsC CP2 ~ C
5 John to teach syntax [on Tuesdays and Thursdays ]F
We can thus conclude that Rooth=s proposal for focus sensitive adverbs such as only can
straightforwardly be carried over to the predicate want. Notice, however, that we have to ensure that
the ~-operator is adjoined to the CP-level of the embedded clause. Other structures, in which the ~-
operator is adjoined to other positions, have to be excluded. The following structure, for example, is
uninterpretable because the elements in C are not of the right type (since ~C is adjoined at the VP
level, the elements in C are of type <e,<s,t>>):
(109) * IP ep
NP VP g eo
Lisa V CP g ri wantsC NP VP
g tu John 5 ~C to teach syntax [on Tuesdays and Thursdays ]F
A structure in which the index C is attached at the IP-level of the higher clause, such as in
(110), leads to a nonsensical interpretation: even though the elements in C are of the right type, C
does not correspond to a set of alternative propositions of the embedded clause. Here, C corresponds
to the following set of alternative propositions: {>Lisa wants John to teach syntax on Tuesdays and
Thursdays=, >Lisa wants John to teach syntax on Mondays, Wednesdays and Fridays}. These are not
the appropriate alternatives that the predicate want requires.
(110)* IP ep
49
IP ~C ep
NP VP g eo
Lisa V CP g ri wantsC NP VP
g 5 John to teach syntax [on Tuesdays & Thursdays ]F
In the following section, I turn to the semantic contribution of the subjunctive mood
morpheme. I propose that the role of the subjunctive mood is to ensure that C contains the appropriate
set of alternatives, or in other words, that the evaluation of contextual alternatives happens in the right
place of the tree, namely at the level of the embedded clause.
7.2.2 The subjunctive mood: evaluation of contextual alternatives
In this section, I return to the Spanish data. I will assume that the subjunctive mood is realized
in a projection above IP, presumably in MoodP12. I will argue that subjunctive mood is an operator
that has the role of the ~-operator, namely to evaluate the alternatives for the matrix predicate. As a
consequence, the evaluation of alternatives necessarily happens in MoodP. Subjunctive mood thus
ensures that the evaluation of alternatives happens at the right place in the tree. Indicative mood, to
the contrary, will be argued to prevent evaluation of alternatives at the level of the embedded clause. I
define the SUBJC-operator as follows:
12 See Bhatt & Yoon (1991) for arguments for the existence of the functional projection MoodP. In proposing that MoodP is between CP and IP, I follow Kempchinsky (1998) and references therein.
50
(111) Definition of SUBJC:
ty g a. 4 SUBJC IP 4o is only defined if g(C ) φ � IP� g
ALT & Card(g(C)) > 1
ty g when defined 4 SUBJC IP 4o = � IP� g
o
ty g b. 4 SUBJC IP 4ALT = {� IP� g
o}
The semantic contribution of the subjunctive mood can best be illustrated with an example
that contains a focused constituent in the embedded clause:
(112) Victoria quiere que Sofía traiga UNA TORTA DE CHOCOLATE.
Victoria wants that Sofía bring:PRES.SUBJ.3SG a cake of chocolate.
>Victoria wants Sofía to bring A CHOCOLATE CAKE.=
I will assume again that the relevant contextual alternatives are the following: >Sofia bring a
chocolate cake=, >Sofia brings ice cream=, or >Sofia brings an apple pie=. The relevant tree structure
corresponding to (112) is given below.
(113) IP2 ri NP VP g ri
Victoria V CP g ri
quiereC C MoodP g ri que SUBJC IP1 5
Sofía traiga [una torta de chocolate]F
Given the definition of the SUBJC-operator, the semantic derivation of (113) is as follows:
(114) � IP1� go = λw. Sofia brings a chocolate cake in w.
51
� IP1� gALT = {λw. Sofia brings a chocolate cake in w, λw. Sofia brings ice cream in w, λw.
Sofia brings an apple pie in w}
� MoodP� go = � SUBJC IP1� g
o is only defined if g(C ) φ � IP1� gALT = {λw. Sofia brings a
chocolate cake in w, λw. Sofia brings an apple pie in w, λw. Sofia brings ice cream in
w}
when defined
� MoodP� go = λw. Sofia brings a chocolate cake in w
� MoodP� gALT = {λw. Sofia brings a chocolate cake in w}
� CP� go = � MoodP� g
o = λw. Sofia brings a chocolate cake in w
� CP� gALT = � MoodP� g
ALT = {λw. Sofia brings a chocolate cake in w}
� wantC� go = λp.λx.λw=.� q: q � p & q 0 g(C ): p >DES x,w= q
� VP� go = � wantC � g
o (� CP2� go) =
= λx.λw=.� q: q � [λw. Sofia brings a chocolate cake in w] & q 0 {λw. Sofia brings a
chocolate cake in w, λw. Sofia brings an apple pie in w, λw. Sofia brings ice cream in
w}: [ λw. Sofia brings a chocolate cake in w] >DES x,w= q
� IP2� go = � VP� g
o ( � NP� go) =
=λw=.� q: q � [λw. Sofia brings a chocolate cake in w] & q 0 {λw. Sofia brings a
chocolate cake in w, λw. Sofia brings an apple pie in w, λw. Sofia brings ice cream in
w}: [ λw. Sofia brings a chocolate cake in w] >DES Victoria,w= q
The resulting meaning can be paraphrased as follows: among the relevant alternatives {>Sofia brings
chocolate cake=, >Sofia brings an apple pie=, >Sofia brings ice cream=} >Sofia brings a chocolate
cake= is the most desirable alternative to Victoria.
I thus propose that MoodP is responsible for the evaluation of contextual alternatives for the
class of predicates that select the subjunctive mood. These predicates require the subjunctive mood in
the embedded clause because they require the presence of a set of contextual alternatives to the
embedded proposition. The SUBJC-operator evaluates this set of contextual alternatives for the matrix
predicate. At the level of the embedded clause, subjunctive mood is thus the morphological
realization of Rooth=s ~-operator.13
13 According to this theory of the subjunctive mood, we expect interaction with other possible focus-sensitive
52
operators in the sentence. For instance, we expect that only should not be able to associate with a chocolate cake in the following example (since we predict that want associates with a chocolate cake): (i) Victoria only wants Sofia to bring A CHOCOLATE CAKE However, as pointed out to me by an anonymous reviewer, in English only can associate with the Focus in the embedded sentence. This may potentially be a problem for the theory. Interaction with other focus-sensititive operators will be left for future research.
53
Let us now examine what would happen if the embedded clause was in the indicative mood. I
will propose that when the SUBJC-operator is replaced by the IND-operator, no evaluation of
contextual alternatives takes place in MoodP. Rather, the ordinary and alternative semantic values of
the IP are simply inherited. In other words, the IND-operator is essentially an identity function.
Is such a definition of the IND-operator sufficient to exclude indicative clauses to appear
under predicates such as want? The IND-operator does not evaluate alternatives that may be
introduced by a focused constituent in the embedded clause, hence the combination of want with an
indicative clause is, in principle, not interpretable (under the assumption that, just like in the case of
only, C has to be constrained by focus and cannot be purely provided by context). However, we still
have to exclude the possibility that the ~-operator could appear in an appropriate position in the
structure and do the work that otherwise would be done by the SUBJC-operator. In other words, we
have to exclude structures of the following kind:
(115)* IP ri NP VP g ri
Victoria V CP g ri
quiereC C MoodP g ri que MoodP ~C
ri IND IP 5
Sofía trae [una torta de chocolate]F
Hence, the IND-operator has to introduce some kind of constraint that prevents evaluation of
alternatives at the level of the embedded clause. Notice that the ~-operator should still be able to
appear in other positions in the tree. This is so since predicates that select the indicative mood also
allow for focused constituents in the embedded clause:
(116) Victoria sabe que SOFIA trajo una torta de chocolate.
Victoria know that Sofía bring:PAST.IND.3SG a cake of chocolate.
>Victoria knows that SOFIA brought a chocolate cake.=
54
In this example, evaluation of contextual alternatives happens at the level of the whole clause. This is
seen from the fact that (116) is felicitous in a context that provides alternatives such as >Victoria
knows that Marcel brought a chocolate cake=, >Victoria knows that Rafael brought a chocolate
cake=. The corresponding tree structure is given in (117).
(117) IP1 ri IP2 ~C ri NP VP g ri
Victoria V CP g ri
sabe C MoodP g ri que IND IP
5 [Sofía]F trajo una torta de chocolate
The following example illustrates that evaluation of contextual alternatives may also happen below
the level of MoodP, for example in the presence of another focus sensitive operator such as only:
(118) Victoria sabe que solo SOFIA trajo una torta de chocolate.
Victoria knows that only Sofía bring:PAST.IND.3SG a cake of chocolate.
>Victoria knows that only SOFIA brought a chocolate cake.=
55
(119) IP1 ri NP VP g ri
Victoria V CP g ri
sabe C MoodP g ri que IND IP2
ri soloC IP3
ri IP4 ~C
5 [Sofía]F trajo una torta de chocolate
We can conclude that for predicates that select the indicative mood the following holds:
evaluation of contextual alternatives is possible but not at the level of the embedded MoodP. In order
to block evaluation of alternatives at the level of MoodP, I thus propose the following definition of
the IND-operator
(120) Definition of INDC:
ty g a. 4 INDC IP 4o is only defined if
If there is a g(C ) φ � IP� gALT then g(C)= {� IP� g
o}
when defined ty g
4 INDC IP 4o = � IP� go
ty g b. 4 INDC IP 4ALT = � IP� g
ALT
This definition of the INDC-operator introduces the presupposition that if there is a contextually
available set of alternatives included in the alternative semantic value of the IP immediately below
MoodP, it only contains one single element, namely the ordinary semantic value of that IP. This
constraint prevents a predicate such as want to combine with an indicative clause: a contextually
available singleton-set is not sufficient for the interpretation of want.
56
The definitions of SUBJC and INDC capture the fact that predicates that select the subjunctive
mood necessarily require the subjunctive mood in the embedded clause: SUBJC evaluates the
contextual alternatives for the matrix predicate, INDC prevents evaluation of the contextual
alternatives for the matrix predicate. Finally, we need to ensure that the predicates that select the
indicative mood also disallow the subjunctive mood in their complement clause. For the moment,
nothing would prevent SUBJC to appear under a predicate that selects the indicative mood. To remedy
for this, I propose that the following constraint holds for the operator SUBJC:
(121) Constraint on SUBJc:
SUBJC can only be licensed if it appears in the scope of a focus sensitive operator.
SUBJC is thus different from the ~-operator: ~C may appear in a sentence even in the absence of a
focus sensitive operator. As it stands, this constraint on SUBJC is a stipulation. However, it captures
well the fact that subjunctive mood is essentially a phenomenon in embedded contexts. Furthermore,
under certain circumstances, predicates that select the indicative mood do in fact allow for the
subjunctive mood in the embedded clause as long as they are embedded under another operator such
as negation or the question operator, as illustrated in the examples below14.
(122) Lucho no dijo que la comida estuviera lista.
Lucho not said that the food be:PAST.SUBJ.3SG ready.
>Lucho didn=t say that the food was ready.=
(123) Dijo Lucho que la comida estuviera lista?
Said Lucho that the food be:PAST.SUBJ.3S ready?
>Did Lucho say that the food was ready?=
14 As pointed out by an anonymous reviewer, the constraint on SUBJC is not precise enough. As it stands, it predicts that SUBJC can be licensed under any focus sensitive operator. Future research will address the question of how the relevant set of focus sensitive operators can be characterized.
57
SUBJC thus can appear under predicates that select the indicative mood, as long as it is licensed by the
presence of negation or the question operator. These are operators that are also known to be focus
sensitive.
To conclude, in this section, I have proposed that the semantic contribution of the subjunctive
mood is to evaluate the contextual alternatives introduced in the embedded clause. To the contrary,
the indicative mood blocks evaluation of contextual alternatives at the level of the embedded clause.
8. Empirical evidence for a semantics of comparison
The proposal defended here makes the following important claim: predicates that select the
subjunctive mood introduce a gradable property (or ordering relation) with respect to which the
contextual alternatives are compared. The ordering relation is contributed by the lexical meaning of
each predicate. Under this assumption, we expect predicates that select the subjunctive mood, just like
gradable adjectives, to appear in certain types of degree constructions. There are only very few
predicates for which this expectation does not hold, namely some of the causative predicates (such as
hacer >make= and lograr >achieve=) and some of the directive predicates (such as ordenar
>order=)15. I proposed earlier, in section 5, that in these cases the comparative meaning component is
more deeply embedded within the meaning of the predicate. These predicates are thus not gradable
predicates as a whole but rather contain a gradable meaning part. As a consequence, we do not expect
them to allow for degree modification on the surface.
In the following, I discuss the distribution of various degree modifiers such as mucho
(>much=), bastante (>quite=), demasiado (>too much=), enormemente (>enormously=), and the
comparative construction with predicates that select the subjunctive/indicative mood. The distribution
of these degree modifiers provides further evidence for the proposal defended here.
8.1 Gradability in the adjectival versus verbal domain
15 See the appendix for more examples.
Gradable predicates such as tall express properties that are ordering inducing: we can compare
individuals according to whether they have the property of being tall to a greater or lesser extent. The
58
crucial difference between gradable and non gradable predicates is that the domains of gradable
predicates can be partially ordered according to some gradient property, whereas the domains of non-
gradable predicates cannot.
For adjectives, modifiers such as >very= are an indicator of their gradable/non gradable
nature: while gradable adjectives can combine with >very=, non gradable ones cannot. The sentence
in (124) expresses that the victim is tall to a high degree, while the sentence in (125) cannot express
that the victim is dead to a high degree:
(124) The victim is very tall.
(125) *The victim is very dead.
Another good test for the gradable/non gradable nature of an adjective is the comparative
construction. Only gradable expressions can appear in comparative constructions:
(126) The victim is taller than his neighbor.
(127) *The victim is more dead than his neighbor.
When we turn to the verbal domain, the tests for gradability are less straightforward. Bolinger
(1972) suggests that we can distinguish degree verbs from non degree verbs by their acceptance of
lexical intensifiers such as quite, rather, so, etc. Thus, in the example (128), the acceptability of the
modifier quite indicates that the predicate exasperate is a degree verb. The modifier quite expresses
that there is a quite high degree of exasperation.
(128) He quite exasperates me. (Bolinger 1972)
However, Bolinger points out that such modifiers do not always signal degree modification. In
fact, a number of these degree modifiers are ambiguous and do not necessarily modify the degree of
the intensity of the property denoted by the predicate. For example, a modifier such as much as in
(129) can also be used to modify the extensibility of a predicate. Here, the modifier indicates that the
quantity of talking was high, not the intensity of talking.
59
(129) He talks too much. (Bolinger 1972)
Doetjes (1997) points out that there is one class of modifiers that is not ambiguous in this way,
but has a much more restricted distribution: high degree adverbs such as >enormously= exclusively
modify the degree of the intensity of a property. The examples below illustrate that enormously can be
used to express a very high degree of appreciation, but not to express a very high quantity of movie-
going.
(130) John appreciated the movie enormously
(131) *Anne goes enormously to the movies. (Doetjes 1997, p.122)
The modifier >enormously= thus seems to be the best test for distinguishing gradable from non-
gradable predicates. I now turn to the distribution of degree modifiers with the predicates under
discussion.
8.2 Degree modification for predicates that select the subjunctive/indicative mood
In this section, I show that predicates that select the subjunctive mood differ from predicates
that select the indicative mood in what type of degree modification they allow.
A wide range of the predicates that select the subjunctive mood in Spanish allow for
modification with enormemente (>enormously=). This holds for all desire predicates, most emotive
factive predicates, all predicates of doubt, all predicates of probability and possibility and some of the
directive and causative predicates, as illustrated in the examples below16.
(132) Marcela desea enormemente que Rafael venga.
Marcela desires enormously that Rafael come:SUBJ.PRES.3SG.
>Marcela enormously wants Rafael to come.=
16 See the appendix for more examples.
60
(133) Sofía se alegra enormemente de que la vengan a visitar.
Sofía SE glad enormously of that PRO her come:PRES.SUBJ.3PL to visit.
>Sofía is enormously glad that they will visit her.=
(134) Te sugiero enormemente que no salgas.
PRO you suggest enormously that PRO not leave:PRES.SUBJ.2SG.
>I suggest enormously that you don=t leave=
However, among the emotive factive predicates, there are some predicates that cannot
combine with high degree modifiers: certain predicates that express high extremes of scales such as es
estupendo (>it is great=) and es extraordinario (>it is extraordinary=), as well as predicates that
express a lower bound of a scale such as aceptar (>accept=), and es suficiente (>it is sufficient=). For
predicates that express such extremes of scales it is not unexpected that they disallow high degree
adverbs. High degree adverbs simply contribute redundant information in this case. Notice that all of
these predicates allow for other degree modifiers such as muy (>very=). Furthermore, as we will see
below, they can also appear in comparative constructions.
There are also a few other emotive factive predicates that do not allow for high degree
modifiers, such as es lógico (>it is logic=), es natural (>it is natural=), es normal (>it is normal=), es
mejor (>it is better=) and es peor (>it is worse=). For some reason, these predicates resist high
degree modification. They do however allow for degree modifiers such as muy (>very=) and mucho
(>much=). Since all the above mentioned examples are adjectives, I will assume that the possibility of
the modifier muy (>very=) is sufficient a test to show that they denote gradable properties. Again, all
of these predicates can also appear in comparative constructions.
To the contrary, none of the predicate classes that select the indicative mood allow for
modification with high degree adverbs such as enormemente (>enormously=) (with the exception of
the class of predicates of certainty). The incompatibility with high degree adverbs is illustrated in the
examples below with an epistemic predicate saber (>know=), and a predicate of communication decir
(>say=).
61
(135) *Sofia sabe enormemente que no puede venir.
Sofia knows enormously that PRO not can:PRES.IND.3SG come.
*=Sofia knows enormously that she cannot come.=
(136) *Alberto dijo enormemente que tenía hambre.
Alberto said enormously that PRO have:PAST.IND.3SG h u n g r y .
* >Alberto said enormously that he was hungry.=
As expected under the proposal, all predicates that select the subjunctive mood (with the
exception of certain directive and causative predicates) allow for modification with a whole range of
other degree modifiers such as mucho (>much=), bastante (>quite=), tanto (>so much=), demasiado
(>too much=):
(137) Marcela teme mucho que Rafael venga.
Marcela fears much that Rafael come:SUBJ.PRES.3SG.
>Marcela very much fears that Rafael will come.=
(138) Sofía se alegra tanto de que la vengan a visitar.
Sofía SE glad so of that PRO her come:PRES.SUBJ.3PL to visit.
>Sofía is so glad that they will come to visit her.=
To the contrary, with predicates that select the indicative mood these modifiers are inacceptable
(when they express modification of the intensity of the property):
(139) *Rafael sabe mucho que no tiene razón.
Rafael knows much that PRO not have:IND.PRES.3SG right.
>Rafael much knows that he is not right.=
(140) *Rafael promete mucho que podrá venir.
Rafael promises much that PRO can:FUT.IND.3SG come.
>*Rafael promises much that he will be able to come.=
62
For predicates that select the indicative mood, when modification with such degree modifiers
is possible, t h e y c a n o n l y e x p r e s s m o d i f i c a t i o n o f t h e
>e x t e n s i b i l i t y
= o f t h e p r e d i c a t e o r
r e p e t i t i o n . T h u s , i n t h e e x a m p l e b e l o w , w i t h t h e p r e d i c a t e o f c o m m u n i c a t i o n d e c i r
(>s a y= )
, d e m a s i a d o c a n o n l y m e a n >
t o o o f t e n=
a n d n o t >
t o o i n t e n s e l y=.
(141) Alberto dijo demasiado que tenía hambre.
Alberto said too-much that PRO h a v e :P A S T . I N D . 3 S G
hunger
>A l b e r t o s a i d t o o o f t e n t h a t h e w a s h u n g r y .
Importantly, for those predicates that can select both the indicative and the subjunctive mood,
degree modification is only possible when the subjunctive mood is selected. This is illustrated here
with the predicate sentir:
(142) Siento enormemente que te hayan hecho daño.
PRO sorry enormously that PRO you have:PAST.SUBJ.3PL done pain.
>I am enormously sorry about the fact that they have hurt you.=
(143) * Siento enormemente que va a haber un problema.
PRO sense enormously that PRO go:FUT.IND.3SG to there-be a problem.
*>I sense enormously that there is going to be a problem.=
The predicate esperar has also been claimed to allow for mood alternation (cf. Ridruejo 1999)17.
When this predicate selects the subjunctive mood it has the meaning of the factive emotive predicate
17 However, the indicative mood is only possible when the embedded clause is about a future event (as pointed out in Ridruejo(1999): (Error! Main Document Only.) Espero que venga /vendrá mi hermano
PRO hope that come:PRES.SUBJ.3SG/come:: FUT.IND.3SG my brother (Error! Main Document Only.) Espero que mi hermano viniera/*ha venido ayer
PRO hope that my brother come:PAST.SUBJ.3SG/*PAST.IND.3SG
63
>to hope=, whereas when it selects the indicative mood it has the meaning >to anticipate=/=to
expect=. Crucially, modification with >enormemente= is possible only when the predicate selects the
subjunctive mood:
(144) Espero enormemente que venga mi hermano
PRO hope enormously that come:PRES.SUBJ.3SG my brother
>I enormously hope that my brother will come=
(145) * Espero enormemente que vendrá mi hermano.
PRO hope enormously that come:FUT.IND.3SG my brother
*>I enormously anticipate that my brother will come.=
Among the predicates that select the indicative mood, an exception is formed by the class of
predicates of certainty such as estar seguro (>be certain=). For these predicates, degree modification
is in fact possible. We will see, however, that the semantics of these predicates does not involve
comparison of contextual alternatives.
Imagine again a context in which there are the following three possibilities: Sofia may bring a
chocolate cake, she may bring an apple pie or she may bring ice cream. In this scenario, the
interpretation of a sentence containing the predicate be certain does not involve ranking the possible
alternatives on a scale of certainty. For example, the sentence in (146) does not imply (147). Rather it
implies (148):
(146) Victoria is certain that Sofia will bring chocolate cake.
(147) Victoria is less certain that Sofia will bring ice cream.
(148) Victoria is certain that Sofia will not bring ice cream.
We have seen previously that for predicates such as want the contextual alternatives are
ranked on a scale of desirability. Thus, if Victoria wants Sofia to bring chocolate cake, this implies
that the other alternatives are ranked lower on the scale of Victorias desirability. As we have seen
here, this is not the case for a predicate such as be certain. If Victoria is certain that Sofia will bring a
chocolate cake, this does not imply that she is less certain about the other alternatives.18
18 The predicate creer >believe= also allows for modification with enormemente and mucho (the only predicate in
64
the class of epistemic predicates).Parallel to predicates of certainty, if Victoria believes that Sofia will bring a chocolate cake, this does not imply that she believes less in the other alternatives. Rather it implies that Victoria believes that Sofia will NOT bring ice cream.
65
Finally, there is one further type of empirical evidence for gradability, namely the kind of
comparative construction illustrated in the example (149).
(149) It is more likely that Sofia will bring a chocolate cake than that she will bring an apple pie.
Notice that, in this example, overt comparison is established between two of the possible alternatives,
namely >that Sofia will bring a chocolate cake= and >that Sofia will bring an apple pie=. This kind of
comparative construction is thus a very straightforward test for the proposal developed here. As
expected, apart from the class of directives and causatives, all predicate classes that select the
subjunctive mood can appear in such comparative constructions. To the contrary, predicates that
select the indicative mood cannot appear in comparative constructions of this kind. This is illustrated
in the examples below.
(150) Sofia teme más que pueda perder su trabajo que no que tenga que trabajar demasiado.
>Sofia fears more that she could loose her job than that she may have to work too much.=
(151) Es más probable que Rafael venga a la fiesta que no que Lucia venga.
>It is more likely that Rafael will come to the party than that Lucia will come.=
(152) *Sofia sabe más que Lucia vendrá a la fiesta que no que Rafael vendrá.
*>Sofia knows more that Lucia will come to the party than that Rafael will come.=
(153) *Sofia comprendio más que tenía que trabajar demasiado que no que podía cambiar de
trabajo.
>*Sofia understood more that she had to work too much than that she could change her job.=
To conclude, in this section I have provided empirical evidence for the claim that predicates
that select the subjunctive mood are gradable predicates. These predicates can combine with degree
modifiers and can appear in a certain type of comparative construction, contrary to the predicates that
select the indicative mood.
66
8.3 Refining the proposal: predicates that select the subjunctive mood have an extra degree
argument
In this section, I will refine the proposal in order to accommodate the empirical facts presented
above. In doing so, I will follow recent semantic literature that has adopted a scalar analysis19 for the
semantics of gradable adjectives. In the scalar approach, gradable adjectives are analyzed as relations
between individuals and degrees. In the literature, there are different variants as to what exactly the
denotation of a gradable adjective should be. I will assume here that a predicate such as tall is a
function of type <d,<e,<s,t>>> :
(154) � tall� = λdd.λxe.λws. x is tall to a degree d in w.
Similarly, I will assume that the predicate want is a function of type <d,<<s,t>,<e,<s,t>>>:
(155) � want� = λd.λp.λx.λw. x wants p to a degree d in w.
The degree morphology then applies to this function and imposes some requirement on the degree.
This requirement is different for each type of degree construction (absolute, comparative, superlative
and equative constructions). For each different construction in which the predicate may appear, this
degree morpheme introduces a different ordering relation. In absolute constructions, the ordering
relation is contributed by an abstract ι-morpheme. In comparative, superlative and equative
constructions the overt morphemes more, less, most or as take the place of this abstract ι-morpheme.
19 See for example Bartsch & Venneman (1973), Bierwisch (1989), Cresswell (1976), Heim (1985, 2000), Kennedy (1999,2001), Kennedy & McNally (2005), Klein (1991), von Stechow (1984a,b). See Kennedy (1999) for an overview of scalar analyses of gradable adjectives.
I will thus propose that, for predicates such as want, the comparison relation is introduced by
an abstract ι-morpheme. Notice that, in such an analysis, the ι-morpheme has the meaning of a
superlative morpheme, rather than an absolute morpheme (as is the case for adjectives). I will not
make any commitments as to what exactly the syntax of the verbal projection should be (see, for
example, Kennedy 1999 who proposes that adjectives have an extended projection DegP that hosts
67
their degree morphology). For the sake of simplicity, I will assume that the degree morpheme is
attached at the V=-level, in the same position in which we find other verbal modifiers. For the
predicate want, this abstract ι-morpheme carries the variable C (which refers to the set of contextual
alternatives to p):
(156) VP ri
V= CP ri p
DegP V ιC want
The denotation of the whole VP should be the following:
(157) � VP� g = λx.λw. � q: q � p & q 0 g(C):
max(λd. x wants p to a degree d in w) > max(λd=. x wants q to a degree d= in w)
where max(S) = ιx:x 0 S & � y[y 0 S Ψ y� x]
We obtain this meaning from the two following parts:
(158) � want� g = λd.λp.λx.λw. x wants p to a degree d in w.
(159) � ιC� g = λP<d,<st,<e,st>>>.λp.λx.λw. � q: q � p & q 0 g(C): max(λd.P(d)(p)(x)(w)) >
max(λd=.P(d=)(q=)(x)(w))
In fact, some clarification is in order to explain how this proposal for the meaning of the
predicate want is related to the definition developed previously in section 4. Before, the comparison
relation >DESα,w was used to compare propositions, here > is used to compare degrees. Based on the
definition of the relation >DESα,w, we can express the meaning of a degree by defining an equivalence
relation (following Cresswell1976):
(160) Definition of an Equivalence Relation .:
p .q iff � z: (p >DESα,w z iff q >DESα,w z) & (z >DESα,w p iff z >DESα,w q)
68
We can now define a relation � >DESα,w between degrees based on the relation >DESα,w. (� >DESα,w) is the set
of all >DESα,w -equivalence classes ( stands for field of a relation: Athe set of all things that are related
in one direction or another to something else@ Cresswell1976, p. 266, and p stands for the degree to
which p is desirable):
(161) p 0(� >DESα,w) iff � p 0 (>DESα,w): p = {z: z .p}
(162) p � >DESα,w q iff p >DESα,w q
The meaning of the predicate want should thus be as follows.
(163) � want� g = λd.λp.λx.λw. x wants p to a degree d in w.
where d 0 (� >DESx,w)
(164) � ιC� g = λP<d,<st,<e,st>>>.λp.λx.λw. � q: q � p & q 0 g(C):
Focus-sensitivity Modification with Modification with overt comparative enormemente mucho/muy construction possible
(intensity) (intensity) recomendar (>recommend=) Υ Υ Υ Υ sugerir (>suggest=) Υ Υ Υ Υ CAUSATIVE PREDICATES hacer (>make=) Υ * * * conseguir (>achieve=) Υ * * * lograr (>accomplish=) Υ * * * evitar (>avoid=) Υ Υ * Υ contribuir (>contribute=) Υ Υ Υ Υ ayudar a (>help=) Υ Υ Υ Υ causar (>cause=) Υ * * * Acknowledgments Over the years, this paper has benefitted from the help and comments of many, many people . I=m especially indebted to Sigrid Beck, Roger Higgins, Angelika Kratzer, Claudia Maienborn, Lisa Matthewson, Barbara Partee, Arnim von Stechow, and Wolfgang Sternefeld, and two anonymous reviewers. I would also like to thank Ana Arregui, Rajesh Bhatt, Nicholas Asher, Jenny Doetjes, Kai von Fintel, Irene Heim, Chris Kennedy, Sveta Krasikova, Doris Penka, Paul Portner, Yael Sharvit, and Philippe Schlenker for many comments and suggestions, as well as the audiences at the SALT X conference at Cornell University, the 30th Symposium of Romance Linguistics at the University of Florida, and the audiences at talks given at the University of Connecticut, at the University of Texas at Austin, at the Université Paris 7, and at the Universität Tübingen. All errors are mine. This paper was written under the support of the SFB 441/B17 project at the Universität Tübingen.